TestiFYI 🧬

Your AI-Powered Evidence Copilot

---

📌 What is TestiFYI?

Imagine being an officer juggling body-cam footage, audio clips, witness statements, and scribbled reports.......all while the clock’s ticking on paperwork and investigations. It’s chaotic, exhausting, and inefficient.

TestiFYI steps in as your AI-powered case co-pilot. It’s a multi-source investigation assistant built to:
🎯 To make sense of scattered evidence and help law enforcement focus more on the field, not the filing cabinet.

TestiFYI lets you upload videos, audios, text notes, and even scanned images — and intelligently processes them to:

• 🎧 Transcribe what was said.
• 🗣️ Detect who said it.
• 👀 Extract key details.
• 🤖 Run powerful LLM-based analysis to flag contradictions or gaps like "Accoridng to the informer the suspect is wearing a blue hoodie, but the person arrested is wearing ablack one"
• 💬 Let officers query the case in natural language like “Where did the robbery happen?”

This project draws inspiration from the incredible work by the folks at Code Four....... a groundbreaking platform that’s revolutionizing how cops handle digital evidence. Their tool auto-generates reports, summarizes videos in real-time, and reduces desk-time for officers by up to 60%. If you haven’t seen their launch post, check it out here. It’s 🔥.

TestiFYI is a smaller, simplified take on what CodeFour is building, with a focus on linking insights across multiple evidence sources and enabling case-level reasoning through LLMs.......all built using open source and free-tier resources! ✌️

---

🛠 Tech Stack

Area	Tool / Library
Frontend UI	streamlit
Audio Transcribe	whisper
Diarization	pyannote.audio, torchaudio
OCR	pytesseract, Pillow
PDF/Text Parsing	PyMuPDF (fitz)
Video Processing	moviepy, pydub
LLM Integration	Groq API with llama-3.3-70b
Backend Utils	Python (json, os, datetime)

---

🚀 Features

• 👮 Upload body-cam video, audio recordings, scanned images, or text files.
• 🧾 Automatic transcription using Whisper & speaker diarization with pyannote.audio.
• 🖼️ Perform OCR on scanned documents like police reports or handwritten notes.
• 📦 Generate a structured JSON aligning speech, speakers, and source metadata.
• 🧠 LLM-powered analysis to uncover inconsistencies or red flags across evidence.
• 💬 Ask natural-language questions and get answers based only on uploaded sources.
• 🎛️ All features accessible via an interactive Streamlit-based UI.

---

📁 Repository Structure

• app.py — Streamlit-based frontend for uploading, analyzing, and querying evidence.
• requirements.txt — All Python dependencies.
• prompts/ — 📄 Prompt templates for LLM interactions
• src/ — 🔧 Core utility logic
  • diarize.py — Speaker diarization with pyannote.audio
  • json_formatter.py — Aligns and saves structured JSON from segments
  • llm_utils.py — Handles Groq LLM interactions with templates
  • ocr_utils.py — Extracts text from scanned images using OCR
  • parse_texts.py — Extracts raw text from .txt and .pdf files
  • transcribe.py — Audio transcription using Whisper
  • video_to_audio.py — Converts video files to WAV audio for transcription

---

Flow 🌱

---

⚙️ Getting Started

1. 🚀 Clone the Repository

git clone https://github.com/Keerthanareddy17/TestiFYI.git
cd testify

2. 🧪 Create a Virtual Environment

python3 -m venv venv
source venv/bin/activate  # On Windows: venv\Scripts\activate

3. 📦 Install Dependencies

pip install -r requirements.txt

4. 🔑 Set Environment Variables

• Create a .env file at the project root and add your Hugging Face Token. You can get one here:

  HUGGINGFACE_TOKEN=your_huggingface_token_here
  

• Set the Groq API key securely in your .env file :

  GROQ_API_KEY=your_groq_api_key_here
  

5. 🖥️ Run the Streamlit App

streamlit run app.py

---

🌟 Interface Overview

📂 Upload Evidence Tab

Upload any combination of supported evidence files:

• Body-cam videos (.mp4, .avi, etc.)
• Audio clips (.wav, .mp3)
• Text documents (.txt)
• Scanned images (.png, .jpg, .jpeg)
• Uploading all types is not mandatory. A single file is enough to proceed.

🧠 Explore & Investigate Tab

• Click 🔍 Run Evidence Analysis
• Let the LLM highlight inconsistencies, contradictions, or suspicious patterns across your evidence.
• Ask natural language case questions like:
  • “What did the witness say about the hoodie?”
  • “Where was the robbery reported to happen?”

⚠️ All responses are generated solely based on your uploaded data.

---

📸 Snapshots of TestiFYI

---

feel free to reach out! 🤝

I'm always up for talking tech, feedback, collaborations, weird LLM behavior stories 😮‍💨

• 📬 Email: katasanikeerthanareddy@gmail.com
• 🌐 LinkedIn